The Impact of AI on Jobs: Differentiation and Diversification Good, Short-Term Displacement Bad

The rise of artificial intelligence (AI) is reshaping industries, sparking debates about job displacement and creation. There is a lot of handwringing about how AI will displace whole sectors and even cause Americans to stop thinking, but the impact is likely to be much more nuanced if history is any guide. The real question is whether the American education and workforce transition system can rise to the challenge.

In 1925, agriculture employed 24.3% of the U.S. workforce (10.7 million people), but now, as defined in 1925 terms, it employs just 1.6% (2.6 million). Huge reduction, right? However, the broader ag ecosystem, including seeds, fertilizers, and ag-tech, has ballooned to 22.1 million jobs and 13.6% of the economy.

In 1925, farming was labor-intensive, with 10.7 million workers toiling on fields. Mechanization—tractors, hybrid seeds, and later biotech—slashed on-farm jobs by 76% by 2023, yet agricultural output soared 10–12x due to productivity gains (total factor productivity up ~2.5x). Meanwhile, new industries like precision ag, bio-fertilizers, and drone tech created 3.5 million jobs in inputs and tech, pushing the sector’s total to 22.1 million. Much less of a premium on labor, much more of a premium (and pay) on human capital.

More people are employed in these fields now than when ag was in its heyday, but they are a lot less as a percentage of the population and they are able to do a whole lot more. The sector has been able to become more specialized, efficient, and productive. Supply chains have multiplied across elements of farming that had not changed in thousands of years (seeds!) and new advances continue to create new fields and new jobs while enhancing the productivity of existing ones. Farming has increasingly become a team sport with a lot of people with complementary skill-sets contributing in unique ways compared to before.

 Manufacturing tells a similar story: direct factory jobs dropped from 14.3% (6.3 million) in 1925 to 7.9% (12.8 million) today, but its ecosystem, including semiconductors and advanced R&D, now supports 28.3 million jobs (17.6% of the workforce). Automation and globalization cut factory jobs’ share, but output jumped ~10x, and new sectors like electronics and aerospace added 4.5 million jobs, with the ecosystem hitting 28.3 million. This “jobless growth” pattern—core jobs shrinking, ecosystem jobs exploding—stems from technological leaps creating new roles upstream (inputs, R&D) and downstream (services, logistics).

AI is likely to cause these and other industries to follow this exact same trajectory. There will be fewer jobs in any specific subset of an industry and more jobs across many more subsets that have not even been invented yet.

Today, direct AI-related roles (e.g., machine learning engineers, data scientists) number 250,000 or so in the U.S. (though climbing rapidly), or less than .1% of the 161 million workforce. These jobs require advanced skills (e.g., PhDs in computer science), and automation tools—like AI writing code or optimizing algorithms—are already reducing demand for mid-tier developers. This mirrors agriculture’s on-farm decline: fewer workers produce more (e.g., GPT-4’s capabilities dwarfed early AI with less human input). But the real story lies beyond the core.

AI is creating roles in:

• Existing industries: energy, particularly nuclear and solar, is booming because of the need to power data centers. Mining and the quest for rare minerals is being reinvented. REITs, real estate, architecture, and construction have also seen new lifts to their established businesses.

• Infrastructure and Inputs: ~1.5 million jobs by 2023 in cloud computing (AWS, Azure), chip manufacturing (NVIDIA, TSMC), and data centers, up 300% since 2015. By 2035, this could double to 3 million as AI demands more GPUs and energy-efficient hardware.

• Applications and Integration: ~2 million jobs in AI-driven sectors like autonomous vehicles (e.g., Tesla’s 50,000+ AI-related roles), healthcare diagnostics (AI imaging), and finance (fraud detection). Integration roles—customizing AI for businesses—added ~500,000 jobs in 2023 alone (e.g., Salesforce AI consultants).

• New Industries: AI ethics, bias mitigation, regulation, and cybersecurity are nascent fields, employing ~100,000 based on the most recent figures available. By 2035, these could hit 1 million, driven by laws like the EU AI Act. Emerging areas like AI-driven synthetic biology or personalized education tech could add another 500,000 jobs.

Total AI-related jobs could reach 10–15 million by 2035 (5–8% of the workforce), up from ~4 million in 2023 (2.5%), and there is no calculating how this will ripple through existing industries and give them new dimensions and areas of opportunity. This reflects agriculture’s shift: core jobs shrink, but the ecosystem—spanning hardware, applications, and new fields—surges.

But for all of the long-term benefits that history suggests will happen, there will be immediate, intense pain for individuals, particularly mid- or late-career employees who may be forced to learn entirely new skills or change industries. There are major social and community effects that techno-optimists may not understand very well.

This will place enormous strain on the nation’s education infrastructure. The old models – basing the school year on the agricultural calendar, teaching children based on assembly line principles, conforming to standardized tests that were heavily influenced by early 20^th century studies of IQ and eugenics – have become straitjackets. The teachers unions have gone from being champions of their members to hide bound artifacts of an earlier era, preserving systems that are increasingly obsolete and serve no one well. I have this visual of an old arthritic man being asked to stretch like a twenty-year-old gymnast. No telling what will pop, but there’s a 100% likelihood something will.

If we want our people to reap more of the benefits and suffer less of the pain of the AI disruption coming ahead, we need an education and workforce transition system that is flexible, responsive, tailored, and efficient, and that supports people throughout their working lives and job and career transitions.

The long-term societal challenge posed by AI is not job loss per se. Done right, more people will be able to do jobs better suited to their gifts and desires.  The question is whether the education and workforce development systems of the US can be remodeled swiftly enough to ease the hard transitions that will occur. If an adaptive education system supporting life-long learning wasn’t a thing before, it is now, because it will be a core strategy for helping our society to cope with the changes that are coming.